path: root/lib/Bytecode/Reader/Reader.h

//===-- Reader.h - Interface To Bytecode Reading ----------------*- C++ -*-===//
//
//                     The LLVM Compiler Infrastructure
//
// This file was developed by Reid Spencer and is distributed under the
// University of Illinois Open Source License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
//  This header file defines the interface to the Bytecode Reader which is
//  responsible for correctly interpreting bytecode files (backwards compatible)
//  and materializing a module from the bytecode read.
//
//===----------------------------------------------------------------------===//

#ifndef BYTECODE_PARSER_H
#define BYTECODE_PARSER_H

#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/GlobalValue.h"
#include "llvm/Function.h"
#include "llvm/ModuleProvider.h"
#include "llvm/Bytecode/Analyzer.h"
#include <utility>
#include <map>
#include <setjmp.h>

namespace llvm {

class BytecodeHandler; ///< Forward declare the handler interface

/// This class defines the interface for parsing a buffer of bytecode. The
/// parser itself takes no action except to call the various functions of
/// the handler interface. The parser's sole responsibility is the correct
/// interpretation of the bytecode buffer. The handler is responsible for
/// instantiating and keeping track of all values. As a convenience, the parser
/// is responsible for materializing types and will pass them through the
/// handler interface as necessary.
/// @see BytecodeHandler
/// @brief Bytecode Reader interface
class BytecodeReader : public ModuleProvider {

/// @name Constructors
/// @{
public:
  /// @brief Default constructor. By default, no handler is used.
  BytecodeReader(BytecodeHandler* h = 0) {
    decompressedBlock = 0;
    Handler = h;
  }

  ~BytecodeReader() {
    freeState();
    if (decompressedBlock) {
      ::free(decompressedBlock);
      decompressedBlock = 0;
    }
  }

/// @}
/// @name Types
/// @{
public:

  /// @brief A convenience type for the buffer pointer
  typedef const unsigned char* BufPtr;

  /// @brief The type used for a vector of potentially abstract types
  typedef std::vector<PATypeHolder> TypeListTy;

  /// This type provides a vector of Value* via the User class for
  /// storage of Values that have been constructed when reading the
  /// bytecode. Because of forward referencing, constant replacement
  /// can occur so we ensure that our list of Value* is updated
  /// properly through those transitions. This ensures that the
  /// correct Value* is in our list when it comes time to associate
  /// constants with global variables at the end of reading the
  /// globals section.
  /// @brief A list of values as a User of those Values.
  class ValueList : public User {
    std::vector<Use> Uses;
  public:
    ValueList() : User(Type::VoidTy, Value::ArgumentVal, 0, 0) {}

    // vector compatibility methods
    unsigned size() const { return getNumOperands(); }
    void push_back(Value *V) {
      Uses.push_back(Use(V, this));
      OperandList = &Uses[0];
      ++NumOperands;
    }
    Value *back() const { return Uses.back(); }
    void pop_back() { Uses.pop_back(); --NumOperands; }
    bool empty() const { return NumOperands == 0; }
    virtual void print(std::ostream& os) const {
      for (unsigned i = 0; i < size(); ++i) {
        os << i << " ";
        getOperand(i)->print(os);
        os << "\n";
      }
    }
  };

  /// @brief A 2 dimensional table of values
  typedef std::vector<ValueList*> ValueTable;

  /// This map is needed so that forward references to constants can be looked
  /// up by Type and slot number when resolving those references.
  /// @brief A mapping of a Type/slot pair to a Constant*.
  typedef std::map<std::pair<unsigned,unsigned>, Constant*> ConstantRefsType;

  /// For lazy read-in of functions, we need to save the location in the
  /// data stream where the function is located. This structure provides that
  /// information. Lazy read-in is used mostly by the JIT which only wants to
  /// resolve functions as it needs them.
  /// @brief Keeps pointers to function contents for later use.
  struct LazyFunctionInfo {
    const unsigned char *Buf, *EndBuf;
    LazyFunctionInfo(const unsigned char *B = 0, const unsigned char *EB = 0)
      : Buf(B), EndBuf(EB) {}
  };

  /// @brief A mapping of functions to their LazyFunctionInfo for lazy reading.
  typedef std::map<Function*, LazyFunctionInfo> LazyFunctionMap;

  /// @brief A list of global variables and the slot number that initializes
  /// them.
  typedef std::vector<std::pair<GlobalVariable*, unsigned> > GlobalInitsList;

  /// This type maps a typeslot/valueslot pair to the corresponding Value*.
  /// It is used for dealing with forward references as values are read in.
  /// @brief A map for dealing with forward references of values.
  typedef std::map<std::pair<unsigned,unsigned>,Value*> ForwardReferenceMap;

/// @}
/// @name Methods
/// @{
public:
  /// @returns true if an error occurred
  /// @brief Main interface to parsing a bytecode buffer.
  bool ParseBytecode(
     volatile BufPtr Buf,         ///< Beginning of the bytecode buffer
     unsigned Length,             ///< Length of the bytecode buffer
     const std::string &ModuleID, ///< An identifier for the module constructed.
     std::string* ErrMsg = 0      ///< Optional place for error message 
  );

  /// @brief Parse all function bodies
  bool ParseAllFunctionBodies(std::string* ErrMsg);

  /// @brief Parse the next function of specific type
  bool ParseFunction(Function* Func, std::string* ErrMsg) ;

  /// This method is abstract in the parent ModuleProvider class. Its
  /// implementation is identical to the ParseFunction method.
  /// @see ParseFunction
  /// @brief Make a specific function materialize.
  virtual bool materializeFunction(Function *F, std::string *ErrMsg = 0) {
    LazyFunctionMap::iterator Fi = LazyFunctionLoadMap.find(F);
    if (Fi == LazyFunctionLoadMap.end()) 
      return false;
    if (ParseFunction(F,ErrMsg))
      return true;
    return false;
  }

  /// This method is abstract in the parent ModuleProvider class. Its
  /// implementation is identical to ParseAllFunctionBodies.
  /// @see ParseAllFunctionBodies
  /// @brief Make the whole module materialize
  virtual Module* materializeModule(std::string *ErrMsg = 0) {
    if (ParseAllFunctionBodies(ErrMsg))
      return 0;
    return TheModule;
  }

  /// This method is provided by the parent ModuleProvde class and overriden
  /// here. It simply releases the module from its provided and frees up our
  /// state.
  /// @brief Release our hold on the generated module
  Module* releaseModule(std::string *ErrInfo = 0) {
    // Since we're losing control of this Module, we must hand it back complete
    Module *M = ModuleProvider::releaseModule();
    freeState();
    return M;
  }

/// @}
/// @name Parsing Units For Subclasses
/// @{
protected:
  /// @brief Parse whole module scope
  void ParseModule();

  /// @brief Parse the version information block
  void ParseVersionInfo();

  /// @brief Parse the ModuleGlobalInfo block
  void ParseModuleGlobalInfo();

  /// @brief Parse a symbol table
  void ParseSymbolTable( Function* Func, SymbolTable *ST);

  /// @brief Parse functions lazily.
  void ParseFunctionLazily();

  ///  @brief Parse a function body
  void ParseFunctionBody(Function* Func);

  /// @brief Parse the type list portion of a compaction table
  void ParseCompactionTypes(unsigned NumEntries);

  /// @brief Parse a compaction table
  void ParseCompactionTable();

  /// @brief Parse global types
  void ParseGlobalTypes();

  /// @brief Parse a basic block (for LLVM 1.0 basic block blocks)
  BasicBlock* ParseBasicBlock(unsigned BlockNo);

  /// @brief parse an instruction list (for post LLVM 1.0 instruction lists
  /// with blocks differentiated by terminating instructions.
  unsigned ParseInstructionList(
    Function* F   ///< The function into which BBs will be inserted
  );

  /// Convert previous opcode values into the current value and/or construct
  /// the instruction. This function handles all *abnormal* cases for 
  /// instruction generation based on obsolete opcode values. The normal cases 
  /// are handled by the ParseInstruction function.
  Instruction* handleObsoleteOpcodes(
    unsigned &opcode,   ///< The old opcode, possibly updated by this function
    std::vector<unsigned> &Oprnds, ///< The operands to the instruction
    unsigned &iType,    ///< The type code from the bytecode file
    const Type* InstTy, ///< The type of the instruction
    BasicBlock* BB      ///< The basic block to insert into, if we need to
  );

  /// @brief Parse a single instruction.
  void ParseInstruction(
    std::vector<unsigned>& Args,   ///< The arguments to be filled in
    BasicBlock* BB             ///< The BB the instruction goes in
  );

  /// @brief Parse the whole constant pool
  void ParseConstantPool(ValueTable& Values, TypeListTy& Types,
                         bool isFunction);

  /// @brief Parse a single constant pool value
  Value *ParseConstantPoolValue(unsigned TypeID);

  /// @brief Parse a block of types constants
  void ParseTypes(TypeListTy &Tab, unsigned NumEntries);

  /// @brief Parse a single type constant
  const Type *ParseType();

  /// @brief Parse a string constants block
  void ParseStringConstants(unsigned NumEntries, ValueTable &Tab);

  /// @brief Release our memory.
  void